Technical Factors behind Discordant Results with Liquid Biopsies

Technical factors are a major source of variation in liquid biopsy test results, according to a study published online March 14 in the Journal of Clinical Oncology.

While biological factors, including low rates of tumor shedding and tumor heterogeneity are known to complicate analysis, the authors highlight that technical factors may be an underappreciated source of testing errors and raise "clinical concern" for false positive (FP) and false negative (FN) tests results.

"Although our study demonstrated that the majority of tumor-plasma discordance is a result of technical factors, with continuous improvement over time, NGS should approach the state of genotyping technology in which the majority of discordance is attributable to biologic factors such as tumor heterogeneity and clonal hematopoiesis of indeterminate potential," write the authors led by Daniel Stetson, from AstraZeneca in Boston, Mass.

The need for a noninvasive test to monitor a patient's ongoing mutational status is fueling the integration of liquid biopsy testing into clinical practice. The U.S. Food and Drug Administration approved several plasma-based genotyping tests for clinical use, including tests that analyze hot spot mutations or specific genes such as EGFR and KRAS, as well as some gene panel tests—Oncomine (Thermo Fisher Scientific), Memorial Sloan Kettering–Integrated Mutation Profiling of Actionable Cancer Targets, and FoundationOne (Foundation Medicine).

Previous studies have shown substantial discordance of genomic results between tumor- and plasma-based results. This discordance has been attributed to low tumor content in plasma, limits of assay detection, and tumor heterogeneity.

The present study sent baseline plasma, tumor, and normal tissue samples from 24 patients to four circulating tumor DNA (ctDNA) sequencing vendors. Variant results were compared between the sources. The samples were considered "challenging" as they had limited ctDNA from early-stage cancers (lung, breast, ovary, and prostate cancers). The four laboratories were all CLIA-certified, commercial NGS laboratories. Vendor B had the smallest panel (approximately 20 genes), whereas the three other vendors had panels of 60 to 70 genes. Vendor names were blinded to encourage communication about laboratory and bioinformatic practices.

"The set of plasma samples analyzed allowed ctDNA NGS methods to be evaluated under challenging conditions … however, the approach may have resulted in a higher-than-expected discrepancy rate," explain the authors. "According to our results and those of others, application of ctDNA assays to early detection analyses may be confounded by FN and FP rates."

The researchers found that all vendor assays performed well above 10 percent variant allele frequency (VAF). However, below 1 percent VAF, most variant calls were discordant. These discordant calls were mainly the result of low-VAF calls, mutational biases, and novel somatic variants, the authors concluded. Additional, technical artifacts included background noise, bioinformatic filtering thresholds, and germline variant calls. The authors suggest that flagging and soft filtering for low-VAF calls, mutational biases, and novel somatic variants may reduce FPs.

Overall, positive predictive value (PPV) against tissue ranged from 36 percent to 80 percent. However, three vendors achieving a PPV of 100 percent, when limited to mutations called at an allelic fraction (AF) greater than 1 percent. For mutations called with an AF less than 1 percent, PPVs were as low as 17 percent.

"Variants detected at less than 1 percent AF are routinely reported by each vendor, and such sensitivity is advertised as a unique strength of plasma NGS assays," the authors note. "Fifty percent (22 of 44) of all true positive somatic variants had a VAF of less than 1 percent, underscoring the importance of analytically validated assays with sensitivity below 1 percent VAF."

Most FP calls were novel variants not found in somatic variant databases. The researchers examined raw data to better understand FN variant calls. One vendor called the PIK3CA H1047R variant, which was missed in another vendor's report even though it was present in that vendor's raw data. Despite similar number of reads for the two vendors, it was seemingly below one vendor's bioinformatics calling threshold. One vendor missed TP53 V143M due to elevated background noise in its assay, but the variant was present and reported by another vendor. In another instance, BRCA2 125I8V was reported by one vendor at 20 percent VAF but was not reported by another vendor despite its presence in that vendor's raw data, possibly because of a bioinformatic filtering of suspected germline variants.

"Plasma cell-free DNA can be a scarce specimen, which makes paired analysis across multiple laboratories extremely challenging," writes lead author Cloud Paweletz, from Dana-Farber Cancer Institute in Boston, Mass., in an accompanying editorial. "Additional investment in validated reference materials could be one step toward establishing a reference point that laboratories can use to confirm the accuracy of their results, as well as a step toward improving the quality of testing for our patients with cancer."

Takeaway: Technical factors, including mutational bias, bioinformatics filtering, and assay sensitivity issues, contribute heavily to discordance in liquid biopsy test results more so than biological factors.